The contribution of remote sensing to the assessment of...

Ann. For. Sci. 63 (2006) 579–595 579c© INRA, EDP Sciences, 2006DOI: 10.1051/forest:2006045

Review

The contribution of remote sensing to the assessmentof drought effects in forest ecosystems

Michel Da*, Dominique Gb, Hervé Jc, Nicolas Sd,Anne Je, Olivier Hc

a ENGREF, UMR Territoires, Environnement, Télédétection et Information Spatiale, Cemagref-CIRAD-ENGREF,500 rue JF Breton, 34093 Montpellier Cedex 5, France

b INRA, Unité de Recherche Écologie fonctionnelle et Physique de l’Environnement, BP81, 33883 Villenave d’Ornon, Francec CNES, 18 avenue Edouard Belin, 31401 Toulouse Cedex 9, France

d Inventaire Forestier National, 32 rue Léon Bourgeois, 69500 Bron, Francee Office National des Forêts, Sylvétude Lorraine, 5 rue Girardet, 54052 Nancy Cedex, France

(Received 28 November 2005; accepted 10 July 2006)

Abstract – Due to their synoptic and monitoring capacities, Earth observation satellites could prove useful for the assessment and evaluation of droughteffects in forest ecosystems. The objectives of this article are: to briefly review the existing sources of remote sensing data and their potential to detectdrought damage; to review the remote sensing applications and studies carried out during the last two decades aiming at detecting and quantifyingdisturbances caused by various stress factors, and especially those causing effects similar to drought; to explore the possibility to use some of thedifferent available systems for setting up a strategy more adapted to monitoring of drought effects in forests.

drought / forest / remote sensing / satellite

Résumé – Contribution de la télédétection à l’évaluation des effets de la sécheresse sur les écosystèmes forestiers. Grâce à leurs capacités desurveillance continue, les satellites d’observation de la Terre pourraient s’avérer utiles pour l’évaluation des effets de la sécheresse sur les écosystèmesforestiers. Les objectifs de cet article sont : de passer en revue rapidement les sources actuelles de données de télédétection et leur potentiel pourla détection des dommages dus à la sécheresse ; de passer en revue les études et applications de télédétection conduites pendant les deux dernièresdécennies et visant à détecter et quantifier les perturbations induites par différents facteurs de stress, et en particulier ceux causant des effets semblablesà ceux de la sécheresse ; d’explorer la possibilité d’utiliser certains des systèmes disponibles pour définir une stratégie adaptée au suivi continu deseffets de la sécheresse sur les forêts.

sécheresse / forêt / télédétection / satellite

1. INTRODUCTION

Earth observation satellites have been used for more than30 years for land cover mapping and forest monitoring. Mostof platforms have been developed by state-owned space agen-cies. Commercial systems with very high resolution capabil-ities, mainly in the optical domain, have been developed foraddressing specific markets, e.g. urban mapping, rapid map-ping after natural disasters and defence needs.

In 2005, more than 60 Earth observation satellites are inoperation and are providing relevant information of the planetenvironment, about half of them carrying dedicated sensorsfor land and vegetation observation at different resolution andspectral capabilities [17]. This wide range of Earth observationsystems offers in principle large possibilities for forest appli-cations, but leads at the same time to specific problems on datacompatibility, calibration, geometry and continuity.

* Corresponding author: [email protected]

No Earth observation system is fully dedicated to monitorand quantify the impact of extreme climatic situations such asthe severe heat and drought of 2003 and a very limited litera-ture on such situations is available in temperate climate, espe-cially in Europe, specifically on drought effects.

The aims of this article are:(1) To briefly review the existing sources and useful physical

principles of remote sensing. The observable biophysicalvariables and processes are presented.

(2) To review the remote sensing applications and studies car-ried out during the last two decades aiming at detecting andquantifying disturbances caused by various stress factors.The potential use of earth observation data for detectingdrought effects can be, with some limitations, derived fromthe similarity of the detected changes with those caused bydrought.This part gives an overview of the state of the art inthe use of remote sensing for detecting and monitor-ing forest changes and drought effects. The first section

Article published by EDP Sciences and available at http://www.edpsciences.org/forest or http://dx.doi.org/10.1051/forest:2006045

http://www.edpsciences.org/foresthttp://dx.doi.org/10.1051/forest:2006045

580 M. Deshayes et al.

outlines the capability of remote sensing to detect andtrack rapid vegetation structure changes such as clear cut-ting or storm damages. Fire-related disturbances, a veryimportant and specific issue in forest management, are notconsidered here. The following sections deal with monitor-ing of changes resulting from continuous and progressivemechanisms, such as forest decline or phenological distur-bance or productivity reduction, with a focus on vegetationanomalies due to drought and water stress. The last sectionaddresses the future prospects given by the process-basedmodels of carbon and water fluxes. The main findings ofthe few papers dealing the severe 2003 drought are pre-sented.

(3) To explore the possibility to use some of the different avail-able systems for setting up a strategy more adapted to mon-itoring of drought effects in forests.

2. BRIEF REVIEW OF THE EXISTING SOURCESOF USEFUL DATA AND OF THE OBSERVABLEBIOPHYSICAL VARIABLES

The multiplication of space technologies, e.g. optical toradar sensors, active and passive systems, is opening new pos-sibilities for deriving some key biogeophysical parameters offorest ecosystems. However, various factors are still limitingresearch advances, e.g. the difficulty in modelling the signatureof tree canopy captured by space sensors, the complexity offorest ecosystem functioning, and the still limited capabilitiesof space observation before reaching an operational dimen-sion. Ground measurements remain mandatory for bringing acomprehensive and consistent picture of forest conditions.

2.1. Remote sensing in the optical domain

In the short wavelengths ranging from visible to infrared(400–2500 nm) the sensors measure the solar radiation re-flected by the Earth surface The ratio between reflected energyon incident energy is called reflectance: expressed as a per-centage, it depends on wavelength and on the way the radia-tion interacts with objects. The processes of reflection, absorp-tion, transmission differ strongly according to the wavelengthrange: visible between 400 and 700 nm (VIS), near infra redbetween 700 and 1100 nm (NIR) and short wave infrared be-tween 1100 and 2500 nm (SWIR).

In far infrared or thermal infrared (TIR) ranging from 8 to14 µm the sensors measure the radiation emitted by the Earthsurface. The surface temperature retrieved from thermal in-frared measurements is determined by energy budget at thesurface.

2.1.1. Spatial resolution of the sensor

It can vary from tens of centimetres (aerial photographs)to several kilometres (meteorological satellites) (Fig. 1). Thespectral response or reflectance of the observed unit on the

ground is an aggregate of spectral responses from differentobjects, e.g. single trees or tree canopies, soil, other vegeta-tion layers, all both in sunlight and in shadow. The larger thesize of the “pixel”, the more there are different objects in-cluded. In medium scale (1:30 000) to large scale (1:5 000)photographs, and in very high resolution satellites images (be-low meter resolution), a tree is covered by several pixels, en-abling detailed canopy structure and texture to be extracted([21] for instance). In high resolution satellite images (10 to100 m), a pixel covers several trees: this resolution is partic-ularly adapted for monitoring forests at stand level. Mediumand low resolution satellite data are well suited for regionalforest surveys and monitoring. Pixels between 100 and 500 mcover one to several hectares, but still contain relevant infor-mation on forest canopy properties. This is still valid for lowerresolution, e.g. 1 km, imagery in widely afforested regions.

2.1.2. Spectral bands and spectral resolution

Space-borne optical imagers can usually operate in thepanchromatic mode (large spectral band width at high spa-tial resolution), and in the multispectral mode (several spec-tral bands with narrower width at lower spatial resolution).The spectral band is characterised by the wavelength and theband width. The finer the spatial resolution, the less is the en-ergy received at sensor level. For technical reasons, it is dif-ficult to develop very sensitive sensors with narrow spectralbands. This is the reason why the spatially finest sensors (be-low meter) operate in the visible domain with a large (about400 nm) panchromatic mode: this source of data is particularlyadapted for detecting structural patterns or features of the for-est stands: limits of different forest types, logging roads, clearcutting, canopy texture, etc. The multispectral mode is moreadapted for characterising vegetation: canopy density, photo-synthetic activity, water stress, fire activity, etc. The numberof spectral bands of a multispectral sensor ranges from just afew (i.e. SPOT satellites) to more than 200 bands on hyper-spectral spectrometers. Spectral bands in the visible (blue tored wavelengths), near infrared (NIR) and short wave infrared(SWIR) are particularly interesting for vegetation monitoring.The bandwidth of satellite sensors is generally around 100 nmbut some low or medium resolution sensors such as MODIS(500 and 1000 m) or MERIS (350 m) present narrower bands(10 to 30 nm) more adapted to the retrieval of certain biophys-ical features. The thermal infrared domain (TIR) is used forstudying water fluxes between vegetation and atmosphere, forestimating the evapotranspiration of vegetation canopies andfor detecting water stress. Several TIR spectral bands are nec-essary for separating temperature and emissivity and for cor-recting atmospheric effects.

All disturbing effects of the signal, e.g. atmospheric in-fluence or directional effects, need to be properly corrected.Imaging sensors with high signal to noise ratio are now con-sidered as a prerequisite for better addressing vegetation pro-cesses. For instance this point is crucial for the reflectance offorest canopies in the visible wavelength which is usually lowespecially in the case of coniferous trees.

Remote sensing and forest drought assessment 581

Figure 1. Spatial resolution and updating frequency for complete coverage of whole Europe associated with presently operating remote sensingsatellites (SVAT = Soil-Vegetation-Atmosphere Transfer, VGT = Vegetation).

2.1.3. Temporal resolution

The revisiting frequency over the same area depends onsatellite and sensor specifications, i.e. sensor swath and sys-tem manoeuvring capability. The temporal resolution rangesfrom 15 min for geostationary satellites to more than 20 daysfor some low orbit satellites (Fig. 1). The along or across trackviewing facility increases the agility of the satellites, givingmore opportunity to capture images of a given site, and thusimproving the temporal resolution. Appropriate algorithmshave been developed for properly correcting long time seriesand for deriving the temporal reflectance with cloud screening,atmospheric correction, geometric correction and radiometriccalibration accounting for directional effects [45] .

For instance, the VEGETATION instrument on SPOT4 and5 satellites, operating at 1.1 km resolution with 2000 kmswath, is covering the entire continents almost every day and isused for studying vegetation processes at small scale, with itsevolution and variation between seasons or years. On the otherhand, the SPOT 5 HRG sensor with its moving mirrors cantake a 60× 60 km image at 2.5 m resolution only every 6 daysover a fixed site: in this case, the images will be taken underdifferent viewing angles. The revisit capability is only 26 daysfor an image taken under the same viewing conditions. Thefrequency of observation with the HRVIR sensor on board ofSPOT4 is similar, but the spatial resolution of HRVIR is lower(10 or 20 m).

2.1.4. Physical processes and observable biophysicalvariables

The reflectance of a forest canopy is related to the wave-length and depends on several biophysical parameters such ascrown closure, Leaf Area Index (LAI), chlorophyll and watercontent of the leaves, architecture of the branches and leaves,

structure and composition of the under-storey layers (bush,forest litter, bare ground, etc.) and sub-surface properties ofsoil. The topography, which is too often neglected, also has anindirect effect, as do the measurement conditions (incidenceangle of the sun rays and of the space sensor, fractions of di-rect and diffuse radiation).

At leaf level, it is well known that radiation is subject toa strong absorption due to chlorophyll pigments in the VISwavelengths, to a strong diffusion controlled by the internalstructure of the leaf in the NIR and by a relatively strong ab-sorption by water in the SWIR. As a consequence forest pa-rameters derived from these driving forces such as LAI andequivalent water thickness do play a role at stand level, to-gether with other ones such as tree cover fraction and to someextent soil moisture.

Radiation absorption in the SWIR is less intense than inthe visible range for typical green leaves and therefore the re-flectance of tree canopies in the SWIR is greater than in thevisible range and lower than in the NIR. Similarly the SWIRband is much more sensitive to variations in LAI than the vis-ible range.

The forest stand structure determines the fractions of il-luminated and shadowed elements composing the vegetationlayers (crown, under-storey and soil). Shadowed surfaces re-ceive a diffused radiation which is much weaker in SWIR thanin NIR. This results in darker shadows in the SWIR than in theNIR. Associated to its low sensitivity to atmospheric effectsand its high signal-to-noise ratio, it makes therefore SWIRvery sensitive to variations in the tree canopy structure, e.g.LAI or cover fraction [13, 42]. Thus this spectral band maybe very useful for detecting structure changes such as clearcuts and thinnings [52, 54]. Forest attributes, such as standingvolume, age and tree height, are often correlated one to an-other to some extent. They can be sometimes retrieved fromremote sensing data by inverting reflectance models related to


biophysical parameters or by applying empirical relationships,since they are related to density of vegetation elements (i.e.green LAI) and their spatial distribution.

Water stress can affect the SWIR signal directly since thereis less water in the leaves. It can change also indirectly the veg-etation response in all the wavelengths by inducing anatom-ical changes in the leaves, or altering pigments or reducingLAI [9, 85, 106].

The NDVI (Normalised Difference Vegetation IndexρNIR−ρREDρNIR+ρRED

, [86]) and various other vegetation indices combin-ing reflectance in red wavelenghts (RED) and NIR [5] areclassically used for quantifying such biophysical variables asLAI, biomass or absorbed photosynthetically active radiationor net primary production ([98], among others). The ratioρNIR−ρS WIRρNIR+ρS WIR

[60] is another useful vegetation index since it tendsto saturate less quickly with the LAI or the cover fraction thanNDVI. SWIR-based vegetation indices are also more sensi-tive to the vegetation moisture, but generally they only de-tect important variations (around 50%) in vegetation moisture,well above ordinary variations, around 20% [51]. The resultscould be improved with the use of narrow spectral bands (10 or20 nm instead of 100 nm) as available from aerial hyperspec-tral sensors such as AVIRIS [33] or certain recent lower spatialresolution sensors such as MODIS [23].

The use of thermal infrared data which allow the retrievalof surface temperature is an efficient way for estimating sur-face fluxes. The relationship between temperature and sensi-ble heat flux gives access to latent heat flux (LE), with theknowledge of the energy balance. LE is a component of theenergy budget. It is controlled by availability of soil water,which results from the water balance. This approach has beensuccessfully used by numerous authors to quantify evapora-tion at regional scale from the thermal infrared spectral bandsof satellite sensors such as AVHRR/NOAA or METEOSAT.The instantaneous brightness temperature measured by satel-lite can be combined with land surface processes models ora simplified semi-empirical relationship with the daily evapo-transpiration in order to estimate the daily value of the actualevapotranspiration. The estimates which have an accuracy onthe order of ±1.5 mm are particularly valuable for describingspatial variations of evaporation difficult to obtain from othertechniques [89]. The accuracy in the retrieval of the surfacetemperature is partly depending on the error on surface emis-sivity and the atmospheric correction procedures. The direc-tional effects on the measured brightness temperature is an-other source of error. The satellite systems well suited for theestimation of fluxes at regional scale have a large field of view;for instance the field of AVHRR/NOAA leads to nadir view an-gles ranging from 0 to 55◦. Lagouarde et al. [63] showed overa maritime pine forest that the hot spot effect due to the sensor-sun geometry is important and the variations between verticaland oblique measurements temperatures may reach about 4 ◦Kfor the moderate to large water stress conditions studied. Thereader will find a basic review on the use of thermal infrareddata for estimating heat fluxes in [89].

Fine resolution data provided by aerial photography andspace-borne sensors have proved to be adapted for character-

ising forest condition, from tree level to stand level: forest/nonforest discrimination, mapping of main species types (conifer-ous, broadleaves), tree density to canopy height.

As regards the use of medium and low resolution satellitedata, the most significant advances in the past years have beenachieved with the use of the high temporal frequency capa-bility for analysing and modelling forest ecosystem function-ing, with the retrieval of biophysical parameters such as veg-etation phenology and cycle duration, LAI, fraction of cover,fraction of Absorbed Photosynthetically Active Radiation (fA-PAR), albedo, soil moisture, energy fluxes, water and carbonfluxes, fuel moisture content. . . The thermal infrared range(TIR) is useful for land surface temperature LST and energyfluxes retrieval, leading to water balance and water stress de-tection. Numerous projects are currently being carried out bythe scientific community for developing and validating the es-timation of these parameters.

Abrupt and drastic structure changes, caused by syIvicul-tural practices (clearcuts, thinnings, etc.) or by natural hazards(fire, storm. . . ) can be detected and quantified by space sen-sors, depending on their spatial resolution and the spatial ex-tent of the phenomenon to be observed. More subtle and pro-gressive changes, mostly caused by natural factors, e.g. pestand disease or drought, are often characterised by a modifi-cation of water and chlorophyll content (pigments composi-tion. . . ), by a slow evolution of the morphology of leaves andthe structure of tree crowns, and ultimately by defoliation andtree decline. Such changes are more difficult to detect, monitorand quantify.

As a conclusion, tracking slow vegetation changes, e.g. wa-ter stress due to severe drought and heat, will require bothmedium or low resolution satellite images for monitoring veg-etation and forest canopy evolution on a daily basis and higherresolution images at lower frequency for better characterisingthe properties at tree and stand levels, as well as for disaggre-gating lower resolution pixels.

2.2. Remote sensing in the microwave domain

The spectral domain of microwaves ranges from about 1 cmto 1 m. There are two kinds of observation. The passive sys-tems observe the radiation naturally emitted by the surface.The active systems or Radar emit a radiation and record itsbackscattering by the earth surface.

2.2.1. Radar

Radar (RAdio Detection And Ranging) technology is atechnology that has been developed and used for many yearsby the military sector and has first become available to scien-tists in 1978 (SEASAT satellite). Despite dramatic advancestowards operational applications in forestry, it still needs sig-nificant efforts from the scientific community before reach-ing a level where data can be used on a routine basis. How-ever, this source of information potentially represents a veryinteresting alternative to optical sensors, especially in regions


where cloud cover is hampering the acquisition of good qualityscenes. Radar information is also in many cases a complemen-tary source of information as radar sensor “sees” the object ina very different way than optical sensors.

The basic principle of a radar system is to transmit shortand high energy pulses and to record the quantity and timedelay of the energy backscattered. Usually, the same antennais used for transmission and reception. The radar electromag-netic radiation is characterised by its direction of propagation,amplitude, phase, wavelength and polarisation either vertical(V) or horizontal (H). Real Aperture Radar (RAR) and Syn-thetic Aperture Radar (SAR) are the two types of imagingradar. For space-borne radar, SAR is the most frequently used.The sequence of pulses is processed on SAR systems to syn-thesise an aperture that is much longer than the actual antenna.The nominal azimuth resolution for a SAR is half of the realantenna size. Generally, the resolution achieved is of the orderof 1–2 m for airborne radar, and 10–100 m for space-borneradar systems. New systems reaching 1 m resolution are ex-pected to be launched in a short term.

The radar backscattering coefficient σ0 provides informa-tion about earth surface and is depending on several key fea-tures: (i) radar system parameters, i.e. frequency, polarisa-tion and incidence angle of the electromagnetic radiation, and(ii) surface parameters, i.e. geometric properties of the object,surface roughness and dielectric constant. The radar observa-tion parameters will determine the penetration depth of the mi-crowave into the ground targets, the relative surface roughnessand possibly the orientation of small scattering elements of thetarget.

Different wavelengths, designated by letters, are used in mi-crowave remote sensing. With longer wavelengths, penetrationof the radiation tends to increase. For instance, over a forestcanopy, radiation in X band (with wavelength around 3 cm)will be limited to a few centimetres, while in C band (withwavelength around 6 cm), the waves will go deeper into thecrowns. In L and P band (respectively around 25 and 60 cm),the penetration is going further down to trunks and eventu-ally to the soil. Thus, the information carried by radar radi-ation is closely related to vegetation biomass, depending onthe interaction of the microwaves with different layers of thevegetation canopy. The penetration is also strongly affected bysurface roughness and moisture. Increasing moisture results inincreasing radar reflectivity.

The European ERS-2 satellite, operating in C band and VVpolarisation at 30-m resolution, followed by ENVISAT-ASARwith similar characteristics, and the Canadian Radarsat-1satellite, operating in the same band with HH polarisation at10 to 100 m, are the existing space systems able to procurecomplementary information on forest conditions. AmplitudeC-band radar data are found to be of limited use for mappingforest types and deforestation [81]. This is due to the ratherquick saturation of the signal with forest biomass in this fre-quency, thus preventing the separation of successive stages ofvegetation regrowth.

With the imaging radars operating in longer wavelengths(L-band, possibly P-band in the future), it is possible to pushback the saturation limit [105]. In addition, the HV polarisa-

tion is found to be more sensitive to biomass than VV. In an-other study, three broad classes of regenerating forest biomassdensity were positively distinguished [69]. In their review, Ka-sischke et al. [58] recommend to use multiband and multi-polarisation SAR data for mapping vegetation and for esti-mating forest biomass with better precision than with singlefrequency and polarisation systems.

As regards monitoring damages to forests, radar data havebeen assessed for detecting and mapping burnt areas. Thesestudies have been using C-band data (ERS, Radarsat) and havebeen carried out in boreal regions, in North America [8, 32,47, 57] and in Siberia [77], in the Mediterranean region [34,35], and in tropical regions [62, 90]. Burnt scar mapping hasbeen found possible, which has generally been explained bychanges in soil humidity [47].

As regards soil moisture monitoring, radar may give someinformation for bare soil or for sparse vegetation. In the case ofdense vegetation like European forests the contribution of thesoil in X- and C-band signals is generally too weak because ofthe strong attenuation by the vegetation layer.

In conclusion, the potential of radar for monitoring the ef-fects of drought are yet to be fully explored. SAR in X or Cband could be sensitive to the modification of low leaf biomassat stand level, but the major drawback is the limitation ofground resolution and the lack of continuous time series.

2.2.2. Passive microwaves

Passive microwave sensors measure the natural microwaveemission of the land surface. The brightness temperature mea-sured by the radiometer depends on the emissivity and thesurface temperature. The variations of emissivity provide in-formation on surface soil moisture and vegetation water con-tent, as with TIR imagery. Contrary to TIR sensors passiveradar sensors are insensitive to cloud cover and can thus pro-vide a complementary information. Their spatial resolution isvery low, 10 km to 100 km, but their temporal resolution high,1 to 3 days. Various studies have shown the ability of the pas-sive microwave sensors to monitor surface soil moisture with ahigh temporal frequency. The different soil moisture retrievalapproaches depend on the way vegetation and temperature ef-fects on microwave signal are accounted for [104]. The atten-uation of microwave emission by vegetation is related to itswater content; it may be estimated from green LAI derivedfrom visible and infrared remote sensing data.

2.3. Complementary nature of ground-basedmeasurements

In situ measurements can be combined with airborne andspace borne data for different reasons: (i) calibration and val-idation of methods or models, (ii) temporal or spatial inter-polation of ground observation network; (iii) assimilation intomodels or simulation tools on ecosystems functioning, forestgrowth simulation and prediction of forest production.


Two permanent (long-term) ground-based observation net-works have been established for monitoring the condition ofthe whole European forests with the so-called Level 1 and 2of the European-ICP Forests /EU system. The national sys-tems for inventorying forest resources from National ForestInventories agencies are also useful when permanent networksof sampling plots are used. Various long-term forest experi-ments are achieved with few sites. We can mention particu-larly the continuous measurements of fluxes of CO2, water,radiation. . . (i.e. Eddy Covariance Tower Network) and theintensive monitoring of phenologic stages (phenologic gar-dens for instance), LAI (litter fall measurement in some ICP-Level2 plots for instance), and growth of trees (i.e. dendro-metric data). These data obtained at local scale are valuablefor calibrating the geospatialisation of processes derived fromremote sensing data. The availability on the region under mon-itoring of other information such as, for instance, meteorolog-ical measurements from weather stations networks or maps ofhydrological soil properties is also useful. Mårell et al. [70]give a classification of these facilities and a rough estimate ofthe facilities available in the different European countries.

3. APPLICATIONS FOR MONITORING FORESTCHANGES AND DROUGHT EFFECTS

Disturbances on forest condition can be caused by numer-ous factors driven by human-induced or natural mechanisms.Several factors are most often interacting with each other, ren-dering the diagnostic even more complex. Remote sensingtools have been widely tested for tracking forest changes, tak-ing benefit of the revisit frequency over a given area combinedwith a relatively large coverage capability.

Rapid changes, e.g. clear cutting, fire scars or storm dam-age..., can usually be detected and quantified with a satisfac-tory accuracy as they occur in a limited period of time on thesame stand – several h to few days – and as observations fromspace can provide timely information right after the distur-bance. But the detection capability depends on the intensityand extent of disturbance, and the availability of recent archivefor cross comparison. The degree of persistence is also a keyfactor in the feasibility of remote sensing for detecting rapidchanges.

Changes resulting from continuous and progressive mech-anisms, such as forest decline or phenological alteration orproductivity reduction, are more difficult to detect with space-derived observations. The relatively low intensity of distur-bance requires long term series of observations before depict-ing any sign of disturbance.

3.1. Detection of sudden changes in forest structure

The development of remote sensing methods dedicated todetection of sudden and strong forest structure changes, e.g.clear cut and storm damage assessment, has been rapidly pro-gressing during the last ten years with the increasing need

to define indicators of sustainable management and to imple-ment certification procedures. In addition, the preparation ofthe European programme on Global Monitoring for Environ-ment and Security is expected to lead to the implementation ofoperational services such as the reporting on forest areas andchanges in the framework of the Kyoto protocol.

The resolution of optical data at 10 to 30 m is too broadfor mapping forest types according to most European NationalForest Inventory schemes. These data have however provedto be effective for updating and enriching existing maps. Theoperational use of such data for an annual mapping of theclear felling of Pinus pinaster stands over the 1 million haLandes forest in Aquitaine Region has been clearly demon-strated [54, 55, 93]. Thus since 1999, IFN the French nationalforest inventory agency has been carrying out the assessmentof annual clear cuts from 1990 onwards over the whole Lan-des forest (Fig. 2), using Landsat 5 TM and Landsat 7 ETMsatellite data [93]. The method is based on a change detectionprocedure, followed by a visual inspection of low probabilitypossible clear cuts [30, 54]. The rate of clear cutting by ageclass is afterwards determined by combining the annual clearcut map with ground inventory plots.

In April 2000, IFN used the clear cut mapping method forassessing the damages of the 1999 storm over the northernpart of the Landes massif [92]. A map was produced with 5damage classes (0–20%, 20–40%, 40–60%, 60–80% and 80–100%). In 2002, satellite remote sensing methods have beentested for mapping storm damage in other French regions andunder different local conditions [94, 95]. The study has shownthat change detection protocols together with segmentationtechniques can be applied to satellite images acquired duringlate spring and summer, leading to satisfactory results. Themethod has been applied to Vosges forests in flat and hilly ar-eas (Fig. 3) [95].

3.2. Monitoring forest health and decline

This section gives an overview of the remote sensing toolsdeveloped during the last 10 to 20 years for monitoring foresthealth and decline. Damage to forest health may occur as aresult from short term biogenic aggressions as well as longterm impact of drought and other abiotic factors.

Typical forest decline symptoms are foliage chlorosis(degradation of chlorophyll pigments), foliage loss, degrada-tion of tree crown structure, and tree mortality. Forest declineand dieback can be caused by various factors, such as pestsand diseases, air pollution, or even long term effect of climaticextremes situations (drought, frost. . . ) etc. The causes are of-ten multiple and difficult to identify and separate from eachother.

Aerial photographs at large scale (1:5 000 to 1:10 000, spa-tial resolution < 30 cm), with panchromatic, colour and bet-ter with infrared colour films, have been commonly used overthe past two decades for assessing individual tree crowns andmapping damage areas. As typical examples of earlier studiestriggered by drought effects one can mention the assessment ofthe oak decline in the Tronçais (central France) forest which


Figure 2. Annual clear cut mapping in Landes forest with Landsat TM (period 1990–1999).

Figure 3. Damage map of 1999 storm using satellite and aerial data over Vosges department (5875 km2), France. Left: global view; Right: localzoom.

occurred after the exceptional 1976 drought [82], as was aswell as Pyrenean piedmont [29]. In the early 1990s, the oaksof the Harth forest (Alsace, north-eastern France) underwenta serious decline following the 1989–1991 dry period, and theforest health condition was mapped [78].

The main symptoms are generally progressive crown de-terioration occurring one or several year(s) after the droughtand not short-term drought symptoms such as foliage brown-ing, withering and early fall. More generally, typical droughteffects are quite rare in temperate forests – the symptoms ob-

served during 2003 represent an extreme case – and many ofthe potentially drought triggered symptoms are assessed asdamage of unknown origin.

So far, the most extensive use of aerial photographs in Eu-rope took place during the 1980s, when several campaignswere launched in order to assess “forest decline”, suppos-edly due to air pollution [1, 31, 49, 83, 84] among others).The main investigations have been carried out in Germany,e.g. Black Forest, in Belgium and in France, e.g. Vosges. Thedamage assessment and their mapping were mostly based on


a multi-stage sampling scheme with the use of aerial pho-tographs for stratification. Geostatistics techniques have beenapplied for optimising the sampling design and assessing thespatial errors on the decline intensity estimates [40]. Stan-dardisation and coordination initiatives have been attemptedat regional level by the European Commission [1,48]. Groundmonitoring networks such as the EU/ICP Forests 16 × 16 kmLevel 1 Network offer a complementary and necessary sourceof information: the information can be spatially extrapolatedwith a high sampling design using large scale aerial pho-tographs. As a consequence, the spatial precision of invento-ries is improved, and spatial processes of decline, e.g. spatialepidemiology and relation with environment variables, are bet-ter understood.

The large scale photography has proved its efficiency formonitoring forest damages (Fig. 4). The identification of thespecies and the estimation of the degradation intensity ofcrown structure of each inventoried tree are accurate whenthey are based on the use of three-dimensional informationobtained from a visual interpretation using a stereoscope. Pho-togrammetric techniques were hardly used for locating thetrees or estimating their size. Now the trend is towards re-placing the film with a digital sensor and replacing the tediousconventional visual interpretation with automated image pro-cessing. The present development of automated methods forretrieving the tree or canopy structure from airborne or space-borne digital images with spatial resolution less than the treesize could be profitable (see for instance [46]).

Aerial photographs at smaller scale (1/10 000 to 1/30 000)and satellite data at metric to decametric resolution are wellsuited for forest monitoring at stand level. Numerous studieson air pollution effects and pests and diseases impacts on for-est condition are reported in the literature since 1980 [2, 7, 44,50, 64, 65, 73, 80, 87, 88, 91, 96, 97, 107], among others) andshow that “severe” damage (affecting a “sufficient” number oftrees) can be easily detected, while scattered tree decline isdifficult to see with the limited resolution of space remotelysensed data [6, 24], and without ground assessment. Finally,the feasibility of depicting forest decline is closely dependingon the topography of the study area, on the structure of foreststands, on the date and frequency of data acquisition and onthe spatial resolution of the remotely sensed data.

Important forest defoliation can be easily detected by satel-lite remote sensing. For example, defoliation by gypsy moth(Lymantria dispar) can be mapped and monitored from satel-lite imagery [19, 56]. Two types of techniques can be usedto map defoliated areas or levels of defoliation: firstly by us-ing only one image taken during the defoliation, and photo-interpreting or classifying it; secondly by using two images,one after and one before the attack, and by comparing the twoimages with rating or differencing techniques. Using colourcomposite transparencies, Ciesla et al. [19] have found somelimitations in the assessment of defoliation intensity, inducingcommission errors, and Joria and Ahearn [56] errors due to thepresence of non-forest areas or forest margins on the scenes.SPOT/HRV colour composites were found to take consider-ably less time (5% only) than the interpretation of aerial pho-tos, yet providing similar results [19]. A Landsat TM classifi-

Figure 4. Detailed view of an infrared colour aerial photograph at1:5 000 taken over Harth forest, France (August 1994, spatial resolu-tion about 15 cm). Rectangle indicates the CHS68 plot of the Level 2European ground-based observation network (French RENECOFORnetwork). Oaks are declining and the lime tree understory is at anearly senescence stage (Guyon et al. 1997 [41]).

cation differentiating two levels of defoliation, moderate andsevere, and no defoliation was found to have a 82% agreementwith aerial photography and supplementary ground data.

More recently, massive defoliations caused by gypsy mothwere observed on the oak in the forest of Haguenau in northernAlsace during 1993 and 1994. These defoliations have beenconsidered as a consequence of the 1989–1991 dry period. Thedefoliation intensity was assessed in the field and recordedwithin a GIS database by the local forest managers (ONF,French National Forest Agency). Landsat TM and SPOT HRVdata taken before and after defoliation were used in orderto investigate the capability of satellite data in detecting de-foliations in this area. The change detection method was a5-step approach [25, 30]: (i) radiometric and geometric pre-processing, (ii) relative radiometric normalisation of the im-ages, (iii) computation of the difference image, (iv) analy-sis of radiometric evolution, and (v) threshold classificationand mapping of gypsy moth damage. Results indicate thatin the defoliated areas the reflectance in the NIR range de-creases, while it increases in the VIS domain and even morein the SWIR domain (Fig. 5). An extension of the damage be-tween 1993 and 1994 was noticed, and the comparison with insitu observations has shown that the satellite–based estimatesagree with ground truth (Fig. 6).

Following these encouraging results, the same method hasbeen applied over two French “départements” of westernFrance (Deux-Sèvres and Vienne, total area 12990 km2) formapping the gypsy moth attack that took place during years1992 and 1993 [24]. Defoliation maps have been produced.However, mapping mortality was not possible since the deadtrees were isolated and scattered.


Figure 5. Gipsy moth defoliation mapping using Landsat TM imagery, Haguenau forest, France. Left, extract: colour composite (SWIR channelin red, NIR in green and Red in Blue). Defoliated areas appear in purple. Right, whole forest: difference image between Landsat 1994 andLandsat 1991 (TM 5 – SWIR channel); defoliated areas are in light shades.

Figure 6. Comparison of gipsy moth defoliation maps derived from Landsat TM imagery (left) and ground observations (right). Haguenauforest, Alsace, France.

3.3. Monitoring drought effects on vegetation

The functioning of forest ecosystems results from complexinteractions and exchanges between individual trees, under-growth vegetation, soil and atmosphere, the climatic condi-tions remaining a major driving force in the evolution and bal-ance of forest ecosystems. The short term impact of climaticextreme events such as severe droughts is a more recent is-sue, thus explaining why only a few investigations have beencarried out on this topic.

This section focuses on water stress and vegetation anoma-lies as an immediate response to a severe drought. The mostinnovative results on the drought of 2003 in Europe were ob-tained on these questions.

3.3.1. Effects of water stress on vegetation

Intensive water stress has various ecological and physicalimpacts on vegetation. Several signs are likely to be detectedfrom remote sensing data. The alteration of chlorophyll andleaf pigments, resulting in leaves turning yellow or brown, in-fluences directly the visible range. The diminution of leaf wa-ter content, if strong, may induce an increase of the short waveinfrared reflectance. Stomatal closure and reduced transpira-tion lead to an increase of the thermal infrared response due tothe elevation of leaf temperature and reduced latent heat trans-fer. Water stress can also modify the orientation and the formof leaves and reduce the green LAI; it ultimately can result inan early partial leaves shedding. These manifestations which

are closely comparable with an acceleration of leaves senes-cence concern all wavelengths.

3.3.2. Vegetation condition

The Normalised Difference Vegetation Index NDVI is com-monly used for monitoring vegetation at continental scale withlarge swath sensors like VEGETATION, AVHRR, MODIS orMERIS. Using their daily observation frequency , inter-annualvariations are easily achievable, giving the opportunity to de-tect seasonal anomalies between two situations.

Some specific indices have been used to monitor the ef-fects of drought, such as the Vegetation Condition Index VCIproposed by Kogan [61] over north America from AVHRRdata time series. The VCI quantifies the vegetation greennessanomalies by comparing the NDVI and its maximal and min-imal values observed during the previous years. Drought im-pact in Brazil was monitored following this method [66]. Morerecent studies refined the knowledge on the seasonal sensi-tivity of the relationships between NDVI and meteorological-drought indices based on precipitation [53].

With these low resolution sensors, it is difficult to studyspecific forest types, because the pixel size is often greaterthan the size of the forest stands. Disaggregation techniquescan be applied to low resolution pixels [15]; more detailed in-formation on the forest canopy can then be extracted. In thisway, Maselli [71] has shown using a AVHRR/NOAA long-term data series that the NDVI values of small pines and oaks


Figure 7. Evolution 2002–2003 of vegetation index (NDVI) from VEGETATION sensor, for months of June, July and August (cf. [45]). Bluecolours indicate no major change of vegetation activity between 2002 and 2003. Yellow to red colours indicate a diminution of vegetation activ-ity (less photosynthesis). In August, the effects of drought are particularly visible in southwest to northeast of France. Fires in Var Departmentare also visible.

forests in Mediterranean region have been decreasing for thelast 15 years, as a possible consequence of the diminution ofwinter rainfall.

The short-term effects of the exceptional 2003 drought onvegetation activity were observed over western Europe us-ing VEGETATION data (Fig. 7). In 2003 an important raindeficit lasted from spring to summer, worsening the impactof an exceptional heat during July and August. The deficit wasmore severe in eastern and south-eastern France. The effects ofdrought and heat were visible in forests (foliage yellowing andbrowning, premature defoliation) and even more so on crops(early drying, early harvesting), and an increased number and ahigher intensity of forest fires were also observed. An averageNDVI derived from all images acquired during June, July andAugust 2003 has been computed for each month, and com-pared with the same periods in 2002. The effects of droughtare clearly visible already in June, with an aggravation of thesituation in July and even more in August (Fig. 7).

Lobo and Maisongrande [67] have detailed the analysisover Spain and France by comparing the 1999 to 2003 an-nual profiles of the VEGETATION-derived NDVI. They haveobserved different responses to drought according to two dif-ferent phytogeographic and climatic regions, i.e. oceanic andMediterranean. Negative anomalies of the vegetation index inthe summer 2003 were greater for herbaceous vegetation ofthe oceanic climate region and for deciduous forests. In theMediterranean region, the NDVI was lower than normal, butthe anomalies were less important in absolute value. They alsocompared the NDVI with the difference between total summerprecipitation and total summer potential evapotranspiration, asan estimate of atmospheric water stress. The results indicatethat water stress is a major factor structuring the geographicvariability of NDVI in the region. The phenological anomaliesof NDVI cannot be generalised for all kinds of forests and needsome further in-depth analyses. An example is given by Coretet al. [20] from a series of monthly 20-m spatial resolution im-ages, acquired in 2002 and 2003 with the SPOT HRVIR sensorover a 50 km × 50 km area of south-western France stronglyaffected by the heat wave. A shortening of the 2003 phenolog-ical cycle was observed for the meadows and crops, but theresponse of the deciduous forests of the studied area was not

clear. In the case of the large maritime pine forest of south-western France, which has not very severely suffered from the2003 drought, Guyon et al. [43] pointed out that its impacton the seasonal cycle of the VEGETATION-derived signal de-pends on the nature of undergrowth vegetation. The droughtled to an early onset in August of the autumn decline phaseof the vegetation index PVI (Perpendicular Vegetation Index).But the effect was not marked over canopies with an evergreenunderstorey.

Other earth observation data sources have also been in-vestigated to study the effect of 2003 drought on vegetationactivity. Multiannual time series acquired over Europe from1998 to 2003 with the Sea-viewing Wide Field-of-view Sensor(SeaWiFS) and from January 2003 onwards with the MediumResolution Imaging Spectrometer (MERIS) instrument havebeen analysed by Gobron et al. [37] to assess the state ofhealth of the vegetation in 2003 and 2004, compared to pre-vious years. By having a similar coverage of the visible andNIR domains (respectively eight and fifteen bands), both sen-sors allow the computation of comparable vegetation indices.The similar vegetation indices, MGVI (MERIS Global Vege-tation Index [38]) and SGVI (SeaWiFS Global Vegetation In-dex [36]), are good estimators of the Fraction of AbsorbedPhotosynthetically Active Radiation (FAPAR), an indicatorof the state and photosynthetic activity of vegetation. Gob-ron et al. [37] have shown that the vegetation growth wasaffected as early as March 2003. An experimental surface wet-ness indicator derived from the SSM/I (Special Sensor Mi-crowave/Imager) microwave sensor presents spatial patternsof water deficit matching –with a certain time lag- areas withnegative FAPAR anomalies. Indeed the water stress was foundto precede the vegetation response by up to one month in someplaces. The situation of spring 2004 was compared with pre-vious years to document the recovery of vegetation. In 2004,the situation has returned to normal, suggesting an absenceof observable medium term effect on vegetation at continentalscale.

3.3.3. Soil moisture and vegetation water stress

Soil moisture is a key parameter for studying the water cy-cle and for monitoring vegetation activity. It can be studied


using space sensors operating in passive microwave, with avery low ground resolution (more than 10 × 10 km). SMOS,a space mission in L band (1.4 GHz) to be launched in 2008by ESA, will be fully dedicated to soil moisture monitoringas well as ocean salinity (Wigneron et al. 2003). Encourag-ing results have been achieved with other systems operating atshorter wavelengths (from 5 to 20 GHz), e.g. SMMR [79] orSMMI (see [37] above).

Besides passive microwave sensors, vegetation water stresscan be assessed using remote sensing systems combining si-multaneous measurements in the VIS, NIR and TIR wave-lengths, as available with NOAA-AVHRR or MSG/SEVIRIsensors. From an operational point of view, NOAA-AVHRRpresents several advantages. The main ones are its temporalresolution (four images per day, due to simultaneous operationof 2 satellites) and its good spectral information (visible, near,medium and thermal infrared). The RED and NIR channels(red, 0.58 to 0.68 µm, and near infrared, 0.72 to 1.10 µm) areused to compute vegetation indices which are correlated withgreen biomass and photosynthetic activity. The two thermal in-frared channels (band 4, 10.3 to 11.3 µm, and band 5, 11.5 to12.5 µm) allow the computation of a Surface Temperature cor-rected for atmospheric and emissivity effects [39, 59, 72, 100].The relation between surface-air temperature and vegetationindices can be useful for estimating water deficit (see for in-stance [28]). The spatial resolution of NOAA (1.1 km at nadirto 6 km in oblique view) gives an integration of local varia-tions and provides an average response well adapted to globalscale [68]. SEVIRI data from Meteosat Second Generation(MSG) satellite has proved to be useful for monitoring veg-etation conditions at high temporal frequency (15-min repeatcycle). Vegetation phenology can be monitored though estima-tion of the fraction of absorbed photosynthetically active radi-ation fAPAR and the leaf area index LAI, and water stress isachievable by combining land surface temperature (LST) andsoil moisture, albeit at very low resolution [76].

Using the TIR channel of NOAA-AVHRR, the relationshipbetween LST and water balance has been studied at regionalscale for temperate forest ecosystems. On coniferous forests,especially in the case of the large maritime pine forest in southwestern France, LST is a good indicator of water stress asthe difference between LST and air temperature may reachabout 10 ◦C during strong water stress periods, which couldbe due to the large contribution of the undergrowth vegeta-tion response [28]. Over broadleaved forests, this correlationis not significant as the difference between LST and air tem-perature is low, whatever the water content of soil, becausethe aerodynamic roughness of the tree canopy is high and thehigh tree cover fraction does not allow for a contribution ofthe understory [11]. In the case of the Mediterranean forestsVidal et al. [101] have shown that the NOAA/AVHRR LSTcan be successfully used for estimating the seasonal variationsof the ratio between the latent heat flux, LE, and the poten-tial latent heat flux, LEp, and for detecting in that way canopywater stress situations.

Finally, numerous references are found in the literatureabout fire monitoring and fire risk assessment from space.Many of the research activities have developed and tested wa-

ter stress indices as a possible tool to anticipate fire risk. Cec-cato et al. [16] have shown that water content at leaf level can-not be retrieved from a vegetation stress index. They definedan Equivalent Water Thickness (EWT) which was found tobe one of the factors influencing the signal in the SWIR do-main. A combination of SWIR and NIR bands is necessaryto retrieve EWT at leaf level. Dennison et al. [22] comparedthe same EWT index to a simple index for measuring regionaldrought, the Cumulative Water Balance index CWBI whichcumulatively sums precipitation and reference evapotranspira-tion over a period of time. EWT and CBWI were found to becomplementary for monitoring live fuel moisture. In anotherstudy, Bowyer and Danson [10] have used canopy reflectancemodels to analyse the canopy reflectance sensitivity to severalparameters including EWT, leaf area index (LAI) or fractionof vegetation cover (Fcov). They have shown that if the vari-ations of LAI and Fcov were restricted to a range of valuesrepresentative of a local site (stand level), the sensitivity ofcanopy reflectance to EWT in the SWIR domain was very im-portant, more than 65% of the total sensitivity. In this context,the monitoring of temporal evolution of vegetation water con-tent from space seems possible. However, if the sensitivity ofthe canopy reflectance to vegetation water content is demon-strated, the radiometric quality of the signal registered by thespace borne sensors is still depending on its dynamic rangeand signal-to-noise ratio, both depending on the impact of at-mospheric scattering and bi-directional effects. The questionof the radiometric quality of the data is even more crucial ifthe variation of the vegetation water context itself is narrow,as it has been observed by ground measurements on severalspecies in the Mediterranean basin [99].

3.3.4. Modelling carbon and water fluxes

In relation to the increase of greenhouse gases in the at-mosphere and the related climate change, major efforts arein progress for a better understanding, monitoring and mod-elling of the carbon and water cycles as driving forces ofglobal warming. A lot of remote sensing studies are carriedout to retrieve LAI and its phenological changes, which arekey variables in photosynthesis, transpiration and energy bal-ance. The estimates can be used as input of process-basedmodels or for validating process simulation results. At globalscale, the lengthening of the vegetation activity cycle can bemonitored using historical long time satellite series [74, 109].In broadleaved forests, key phenological stages, e.g. bud-burst, senescence, or length of the seasonal growth, are easilymonitored from space at regional scale [27]. In situ measure-ments tend to confirm the satellite-derived phenological cycle,the latter being only based on detection of seasonal changes inthe remote sensing signal (NDVI), without a need for an ab-solute estimation of LAI. For estimating LAI the situation ismore complex. The saturation of the NDVI with high LAI val-ues makes their retrieval difficult. The used algorithms do notoften account for the mixing of various vegetation structures inlow resolution pixels. This downscaling problem is also crit-ical for the validation of LAI estimates, but also FAPAR orproductivity estimates, which more and more become standard


products available to any user. Wang et al. [102,103] addressedthese questions by comparing on several forest sites local con-tinuous ground measurements of LAI, fAPAR and gross pri-mary production (GPP) with their retrieval from medium orlow resolution satellites (MODIS, VEGETATION, AHHRR).

Carbon and water fluxes can be estimated using functions ofsurface transfer between soil, vegetation and atmosphere [14],and simulations of ecosystem functioning processes [75]. Nu-merous studies address the problem of the coupling of theseprocess models with remote sensing data (see [3]) and theirspatial parameterisation. The assimilation of thermal infraredresponse into the MuSICA model of Ogée et al. [75] is cur-rently being tested over maritime pine stands, and a reductionof errors in the water budget due to the directional and instan-taneous measurement of the surface temperature is particularlyexpected.

Extreme conditions like the 2003 heat and drought eventmodify the functioning of vegetation: high temperatures, oftencombined with a severe shortage of water, reduce the vegeta-tion activity (photosynthesis), and as a consequence LAI andthe fraction of radiation absorbed by plants (fAPAR) are re-duced. This may result in loss of productivity, e.g. for crops.An index of the climatic impact on vegetation production hasbeen developed by Zhang P. et al. [108]: this production in-dex was found to be a good indicator of a drought event asmeasured by meteorological data.

Model predictions of climate changes and temperature in-crease tend to indicate that vegetation should be in morefavourable conditions in most cases, with increased carbon up-take. However, these schemes are not taking into considera-tion extreme conditions such as the heat wave of 2003. Usinga terrestrial biosphere simulation model to assess continental-scale changes in primary productivity, Ciais et al. [18] haveshown that the net ecosystem carbon balance, resulting fromthe difference between the two opposite carbon fluxes GrossPrimary Productivity GPP and ecosystem respiration TER, hasbeen strongly affected in June to September 2003, mainly be-cause of a strong reduction of GPP accompanied by a lesserreduction of TER. The maximum of reduction is observed overFrance and Germany, and to a lesser extent in Italy and Spain.Forest ecosystems in the temperate region have been moresensitive to drought than Mediterranean ecosystems where thevegetation is already adapted to high temperatures and scarcityof summer rain. These results were based on carbon fluxesmodelling using climate and weather data and a land use mapderived from high resolution remote sensing data (i.e. CorineLand cover), The simulations reproduced well the GPP andTER anomalies observed over the CarboEurope Eddy Covari-ance Tower Network. They were also found to be consistentwith a fAPAR map derived from space observations (MODISsensor on board Terra/EOS satellite).

4. CONCLUSION: WHICH OBSERVATIONSTRATEGY FOR DROUGHT EFFECTSMONITORING?

The only existing and operational infrastructure able tomonitor forest condition over a very large extent is the pan-

European ground observation network (ICP Forests / EUlevel 1). If this 16×16 km network is adapted to an assessmentat national or sub-national (regional) scales, it cannot be usedat local scale. An access to information at stand scale is neces-sary for forest management or as an aid to the understandingof spatio-temporal disturbance processes (such as the propaga-tion of bark beetles infestation within and between stands, orthe variations between crown condition change of the ground-observed plot and of its surrounding) In such a case, one prac-tical strategy would be to combine high resolution satellite orairborne observations with those of the existing ground sys-tem, as a complementary source of information for improvingthe geospatialisation of interest variables at local scale. Formapping forest damages, the remote sensing data could beused for Interpolating ground observations with geostatisticstechniques (i.e. co-kriging).

Airborne and spaceborne sensors represent indeed a uniquesource of information for monitoring forest response to the2003 drought at local to regional scale: most of forest canopyanomalies can be easily detected from space, with howevera limitation on the true size of the impact, as damage at treelevel is still difficult to achieve. An adapted strategy is sum-marised in Table I. It proposes to combine low and high res-olution satellite data with in situ data to carry out vegeta-tion monitoring at local to regional (sub-continental) scales.Low resolution satellites, e.g. with a spatial resolution rang-ing from 250 m to several kilometres, are well adapted to acontinuous monitoring of global forest condition, with dailyto hourly revisit frequency, leading to cloud free compositingevery few days. This capability has proved to be extremelyuseful for monitoring forest seasonal and inter-annual activity.The anomalies of vegetation activity, as seen from a vegetationindex or water stress index, could be detected almost in near-real time with VEGETATION, SeaWiFS, MERIS and SEVIRIdata, as early as August 2003. This capability can be extendedto following years in order to analyse the forest response todrought in the long term.

At local scale, finer resolution data are more appropriate forderiving parameters at stand to tree level, but only over siteslimited in extent. The constellation of the three SPOT satellitesmay offer a daily revisit over only few selected sites in West-ern Europe. In the coming years, new systems will be ableto offer enhanced possibilities for monitoring forest stands athigh revisit frequency and with richer spectral information:the VENµS mission, to be launched in 2009, will be able tocapture the same image under the same conditions every twodays, with 12 spectral bands at decametric resolution. This ap-proach will be limited to a sampling scheme with the acquisi-tion of images over sample sites where ground measurementsare available, e.g. LAI, soil moisture, fraction of vegetationcover, water and carbon fluxes. . .

Observations from space are able to offer a decisive addedvalue through the synoptic and comprehensive coverage ca-pability, which is critical for better understanding the spatialvariability of drought effects on forests. The main drawback isthe lack of consistency between the different resolution modes,leading to complex schemes in data fusion or multi-sensor ap-proaches. Due to technical limitations, the finer the resolution,


Table I. Possible Earth observation strategies for drought effects monitoring.

Regional monitoring Local monitoring

Geographical extent Europe 1–10 000 km2, selected sites

Spaceborne data sourceSensors with wide field of view (e.g. VEGETA-TION, MODIS, MERIS, AVHRR) and very highrevisit frequency

High resolution sensors with high revisit frequency(e.g. SPOT constellation, VENµS, Rapid Eye), or lowrevisit frequency (Landsat ETM, ASTER. . . )

Objectives

• Monitoring vegetation condition and phenolog-ical changes: seasonal reflectance and vegeta-tion indices (NDVI) for year 2003 and follow-ing years to be compared with previous yearsas reference

• Estimating vegetation surface parameters (LAI,fAPAR...) and integration (or assimilation) intophysical and physiological processes models :carbon up-take (GPP, NPP), water fluxes

• Land use and forest inventory applications: landuse and forest mapping , anthropogenic changes

• Detecting and mapping aerial dieback effects anddamages: strong changes of forest conditions orcrown conditions, concentrated damages

• Temporal vegetation profiles (LAI, fcover. . . ) In-tegration or assimilation into models of vegeta-tion functioning or of growth at stand level (lim-ited to sites with intensive ground measurements)

Advantages• Data quickly and easily available, free of

charge• Global snapshot of Europe• Temporal profiles of vegetation phenology

• Analysis of forest response at stand level• Useful for disaggregating low spatial resolution

remote sensing data

DisadvantagesAggregation problem due to the low spatial resolu-tion, difficulty to analyse small forest stands

No archive as referencePossibility to monitor only a limited number of sites

Key issue

The two approaches are complementary for monitoring the variability of forest response to drought intime and spaceCoupling with existing ground systems is mandatory: research sites with instrumentation, European mon-itoring networks (ICP level I and level II), Eddy Covariance Tower (Carbon fluxes). . .

the lower the revisit capability: this makes it almost impossi-ble to set up a space observatory over permanent plots withthe double capability in resolution and sufficient temporal fre-quency, e.g. several snapshots per month.

The research efforts carried out so far tend to indicatethat, when monitored at sub-continental (regional) to conti-nental scale, vegetation, and specially grasslands and less sen-sitive Mediterranean vegetation types, has quickly recoveredfrom the 2003 situation. If deciduous forests are found to beseverely affected in some areas, it is at this point difficult toknow how these vegetation types did recover during 2004,2005 and further. It should be stressed that 2005 has been an-other year with water shortage over most of western Franceand the Iberian Peninsula, thus increasing vegetation stress,whilst it complicates the ability to detect the long term effectof the 2003 drought and heat wave.

New lines of research are expected to contribute applica-tions to the monitoring of vegetation conditions and the detec-tion and assessment of anomalies and damage. Among them:

– the development of digital analysis tools adapted to the as-sessment of scattered damage on very fine resolution data;

– the development of strategies for a spatial assessment ofdamages including the use of remote sensing data;

– developments in a number of water stress issues, such asthe identification of potential risk areas [12], its interac-tions with biotic factors [26] and its impact on biodiver-sity [4];

– developments in water budget modeling and in ecosystemmodeling, for their development and their validation, withthe use of phenologic information provided by very hightemporal frequency sensor, together with multiannual in-formation derived from high and very high resolution dataon tree and stand health and vigour at short and long term.

In the medium term (by 2010), the GMES (Global Monitor-ing for Environment and Security) Programme launched bythe European Commission and the European Space Agency isexpected to provide data series more suited to a comprehen-sive monitoring of forest condition at local to regional scales.GMES is based on both space and ground infrastructures. Asfor land environment, the space infrastructure should be ableto deliver cloud free compositing products at decametric res-olution every week. This facility is expected to be highly rel-evant for monitoring forest conditions in Europe. One of themajor advances of GMES will be the setting up of opera-tional services dedicated to an effective surveillance of ourenvironment.

REFERENCES

[1] Anonymous, Remote sensing applications for forest health statusassessment. European Union Scheme on the protection of forestsagainst atmospheric pollution, 2nd ed., Office of Publications ofEuropean Communities, Luxembourg, 2000, 216 p.

[2] Anonymous, SEMEFOR, Satellite based environmental monitor-ing of European forests. Project report. European Commission,


Research Directorate-General Science, Research and develop-ment, Environment and climate program 1994–1998, ContractENV4-CT97-0398, ISBN92-894-0851-0, 2002, 103 p.

[3] Anselmi S., Chiesi M., Giannini M., Manes F., Maselli F.,Estimation of Mediterranean forest transpiration and photosynthe-sis through the use of an ecosystem simulation model driven byremotely sensed data, Glob. Ecol. Biogeogr. 13 (2004) 371–380.

[4] Archaux F., Wolters V., Impact of summer drought on forest bio-diversity, Ann. For. Sci. 63 (2006) 643–650.

[5] Bannari A., Morin D., Bonn F., Huete A.R., A review of vegetationindices, Remote Sens. Rev. 13 (1995) 20–95.

[6] Bazire P., Guyon D., Jolly A., Riom J., Lallemand C., LegendreG., Étude par Télédétection spatiale du dépérissement des forêtsvosgiennes, in: Colloque Intern. SPOT 1, Utilisation des images,bilan, résultats, Paris, 23–27 novembre 1987, pp. 997–1011.

[7] Bochenek Z., Ciolkosz A., Iracka M., Deterioration of forestsin the Sudety Mountains, Poland, detected on satellite images,Environ. Pollut. 98 (1997) 375–379.

[8] Bourgeau-Chavez L.L., Kasischke E.S., Brunzell S., Mudd J.P.,Tukman M., Mapping fire scars in global boreal forests usingimaging radar data, Int. J. Remote Sens. 23–20 (2002) 4211–4234.

[9] Bowman W.D., The relationship between leaf water status, gas ex-change, and spectral reflectance in cotton leaves, Remote Sens,Environ. 30 (1990) 249–255.

[10] Bowyer P., Danson F.M., Sensitivity of spectral reflectance to vari-ation in live fuel moisture content at leaf and canopy level, RemoteSens. Environ. 92 (2004) 297–308.

[11] Bréda N., Duchemin B., Granier A., Lagouarde J.P., Ogée J.,Relationships between surface temperature and fluxes: a compar-ative analysis for temperate deciduous and coniferous forests, in:Proc. International conference, ALPS-CNES, Remote sensing andvegetation productivity, Méribel, 18–22 January 1999, P-15, 4 p.

[12] Bréda N., Huc R., Granier A., Dreyer E., Forest trees and standsunder drought: a review of ecophysiological responses, adaptationprocesses and long-term consequences, Ann. For. Sci. 63 (2006)623–642.

[13] Brown L., Chen J.M., Leblanc S.G., Cihlar J., A ShortwaveInfrared Modification to the Simple Ratio for LAI Retrieval inBoreal Forests: An Image and Model Analysis, Remote Sens.Environ. 71 (2000) 16–25.

[14] Calvet J.-C., Noilhan J., Roujean J.-L., Bessemoulin P.,Cabelguenne M., Olioso A., Wigneron J.P., An interactive vege-tation SVAT model tested against data from six contrasting sites,Agric. For. Meteorol. 92 (1998) 73–95.

[15] Cardot H., Faivre R., Maisongrande P., Random effects varying-time regression models: applications in remote sensing, in: AntochJ. (Ed.), Compstat 2004, Physica-Verlag, 2004, pp. 777–784.

[16] Ceccato P., Flasse S., Tarantola S., Jacquemoud S., Grégoire J.M.,Detecting vegetation leaf water content using reflectance in theoptical domain, Remote Sens. Environ. 77 (2001) 22–33.

[17] CEOS, Committee on Earth observation satellites, http://www.ceos.org/pdfs/CEOS_brochure_Sep04E.pdf. 2004, 12 p.

[18] Ciais P., Reichstein M., Viovy N., Granier A., Ogée J., Allard V.,Aubinet M., Buchmann N., Bernhofer C., Carrara A., ChevallierF., De Noblet N., Friend A.D., Friedlingstein P., Grunwald T.,Heinesch B., Keronen P., Knohl A., Krinner G., Loustau D.,Manca G., Matteucci G., Miglietta F., Ourcival J.M., Papale D.,Pilegaard K., Rambal S., Seufert G., Soussana J.F., Sanz M.J.,Schulze E.D., Vesala T., Valentini R., Europe-wide reduction inprimary productivity caused by the heat and drought in 2003,Nature 437–7058 (2005) 529–533.

[19] Ciesla W.M., Dull C.W., Acciavatti R.E., Interpretation of SPOT-1color composites for mapping defoliation of hardwood forests bygypsy moth, Photogramm. Eng. Remote Sens. 55 (1989) 1465–1470.

[20] Coret L., Maisongrande P., Boone A., Lobo A., Dedieu G., GouauxP., Assessing the impacts of the 2003 hot and dry spell withSPOT HRVIR images time series over south-western France, Int.J. Remote Sens. 26 (2005) 2461–2469.

[21] Couteron P., Pelissier R., Nicolini E.A., Paget D., Predicting trop-ical forest stand structure parameters from Fourier transform ofvery high-resolution remotely sensed canopy images, J. Appl.Ecol. 42 (2005) 1121–1128.

[22] Dennison P.E., Roberts D.A., Thorgusen S.R., Regelbrugge J.C.,Weise D., Lee C., Modeling seasonal changes in live fuel moistureand equivalent water thickness using a cumulative water balanceindex, Remote Sens. Environ. 88 (2003) 442–452.

[23] Dennison P.E., Roberts D.A., Examining seasonal changes incanopy moisture and vegetation recovery from wildfire us-ing AVIRIS time series data, in: AVIRIS Earth Science andApplications Workshop, March 31–April 2, 2004, Pasadena,California.

[24] Deshayes M., Stach N., Cartographie par télédétection des défoli-ations des chênes par le bombyx disparate dans le centre-ouest.Rapport final, Convention DERF N˚ 01.40.14/98, 1999, 29 p.,2 cartes hors-texte.

[25] Deshayes M., Stach N., Malphettes J.B., Utilisation des imagessatellitales pour l’observation des défoliations dues au bombyxdisparate en forêt de Haguenau, Les Cahiers du DSF 1 (1998)87–89.

[26] Desprez-Loustau M.L., Marcais B., Nageleisen L.M., Piou D.,Vannini A., Interactive effects of drought and pathogens in foresttrees, Ann. For. Sci. 63 (2006) 595–610.

[27] Duchemin B., Goubier J., Courrier G., Monitoring phenologi-cal key-stages and cycle duration of temperate deciduous forestecosystems with NOAA-AVHRR data, Remote Sens. Environ. 67(1999) 68–82.

[28] Duchemin B., Guyon D., Lagouarde J.P., Potential and limits ofNOAA-AVHRR temporal composite data for phenology and waterstress monitoring of temperate forest ecosystems, Int. J. RemoteSens. 20 (1999) 895–917.

[29] Durand P., Gelpe J., Lemoine B., Riom J., Timbal J., Le dépérisse-ment du chêne pédonculé dans les Pyrénées Atlantiques, Rev. For.Fr. (Nancy) 35 (1983) 357–368.

[30] Durrieu S., Deshayes M., Méthode de comparaison d’images satel-litaires pour la détection des changements en milieu forestier; ap-plication aux Monts de Lacaune (Tarn, France), Ann. For. Sci. 51(1994) 147–161.

[31] Farcy C., Aerial photography and evaluation of Norway sprucedecline in the Ardennes, Ann. For. Sci. 46 (1989) 155–171.

[32] French N.H.F., Bourgeau-Chavez L.L., Initial observations ofRADARSAT imagery at fire-disturbed sites in interior Alaska,Remote Sens. Environ. 68 (1999) 89–94.

[33] Gao B.-C., NDWI – A Normalized Difference Water Index forRemote Sensing of Vegetation Liquid Water from Space, RemoteSens. Environ. 58 (1996) 257–266.

[34] Gimeno M., San-Miguel-Ayanz J., Evaluation of RADARSAT-1data for identification of burnt areas in Southern Europe, RemoteSens. Environ. 92 (2004) 370–375.

[35] Gimeno M., San-Miguel-Ayanz J., Schmuck G., Identification ofburnt areas in Mediterranean forest environments from ERS-2SAR time series, Int. J. Remote Sens. 25 (2004) 4873–4888.


[36] Gobron N., Mélin F., Pinty B., Verstraete M.M., Widlowski J.-L.,Bucini G., A Global Vegetation Index for SeaWiFS: Design andApplications, in: Beniston M., Verstraete M.M. (Eds.), SatelliteRemote Sensing Data and Climate Model Simulations: Synergiesand Limitations, Kluwer Academic Publishers, Dordrecht, 2001.

[37] Gobron N., Pinty B., Melin F., Taberner M., Verstraete M.M.,Belward A., Lavergne T., Widlowski J.L., The state of vegeta-tion in Europe following the 2003 drought, Int. J. Remote Sens.26 (2005) 2013–2020.

[38] Gobron N., Pinty B., Verstraete M.M., Govaerts Y., The MERISGlobal Vegetation Index (MGVI): description and preliminary ap-plication, Int. J. Remote Sens. 20 (1999) 1917–1927.

[39] Gu X.F., Seguin B., Hanocq J.F., Guinot J.P., Evaluation andcomparison of atmospheric correction methods for thermal datameasured by ERS1-ATSR, NOAA11-AVHRR, and Landsat5-TMsensors, in: Proceedings of ISPRS 6th International SymposiumPhysical Measurements and Signatures in Remote Sensing, Vald’Isère, France, 17–21 Janv. 1994, pp. 793–800.

[40] Guyon D., Riom J., Qualité de l’estimation spatiale d’unphénomène évolutif appréhendé par échantillonnage. Exemple dedépérissement des forêts vosgiennes par photographie aérienne,in: Buche P., King D., Lardon S. (Eds.), Séminaire INRA Gestionde l’espace rural et SIG Florac 22–24 octobre 1991, INRA, 1992,pp. 295–305.

[41] Guyon D., Riom J., Kicin J.L., Letouzé F., Courrier G.,Applications de la télédétection et des systèmes d’informationgéographique à l’étude et la gestion des peuplements forestiersdépérissants. Compte-rendu final, novembre 1997, ProjetCEE/DGVI (règlement 3528/86) N◦ 92.60.fr.0020. 1997, 75 p.

[42] Guyon D., Berbigier P., Courrier G., Lagouarde J.P., Moreau P.,Sensitivity analysis of coniferous forest reflectance with canopystructure and undergrowth characteristics from satellite data andmodelling (case study: Landes maritime pine forest), in: CNES(Ed.), 8th International Symposium Mesures Physiques et signa-tures en télédétection, Aussois, 8–12 January 2001, 387–392.

[43] Guyon D., Cardot H., Hagolle O., Monitoring and mapping thephenology of the maritime pine forest of southwestern Francefrom VEGETATION time-series, in: 2nd International SymposiumRecent Advances In Quantitative Remote Sensing, Valencia(Spain), September 25th–29th, 2006.

[44] Guyon J.P., Guyon D., Riom J., Causes et gestion du dépérisse-ment du pin maritime sur le littoral Nord-Atlantique. Rev. For.Fr. (Nancy), N◦ spécial Les dépérissements d’arbres forestiers –causes connues et inconnues, 46 (1994), 485–494.

[45] Hagolle O., Lobo A., Maisongrande P., Cabot F., Duchemin B.,De Pereyra A., Quality assessment and improvement of temporallycomposited products of remotely sensed imagery by combinationof VEGETATION 1 & 2 images, Remote Sens. Environ. 94 (2005)172–186.

[46] Haara A., Nevalainen S., Detection of dead or defoliated sprucesusing digital aerial data, For. Ecol. Manage. 160 (2002) 97–107.

[47] Harrell P.A., Bourgeau-Chavez L.L., Kasischke E.S., FrenchN.H.F., Christensen N.L. Jr., Sensitivity of ERS-1 and JERS-1Radar Data to Biomass and Stand Structure in Alaskan BorealForest, Remote Sens. Environ. 54 (1995) 247–260.

[48] Haussmann T., Lorenz M., Fisher R., Internal Review of ICPForests, 16th Task Force Meeting of ICP Forests (Gent, Belgium),May 2000, UN/ECE Report, 133 p.

[49] Hildebrandt G., Pilotinventur für eine Europaïsche Waldschadensinventür, in: Proceed. IUFRO Conference, 402/605,Inventoring and Monitoring Endangered Forests, Zürich, 1985,pp. 237–242.

[50] Holmgren P., Thuresson T., Satellite remote sensing for forestryplanning – A review, Scand. J. For. Res. 13 (1998) 90–110.

[51] Hunt E.R. Jr., Rock B.N., Detection of changes in leaf water con-tent using near- and middle-infrared reflectances, Remote Sens.Environ. 30 (1989) 43–54.

[52] Jeanjean H., Deshayes M., Synthèse Zones forestières et SPOT4MIR, in: SPOT4 MIR Synthèse thématique, Centre Nationald’Études Spatiales, Paris, France, 1998.

[53] Ji L., Peters A.J., Assessing vegetation response to drought inthe northern Great Plains using vegetation and drought indices,Remote Sens. Environ. 87 (2003) 85–98.

[54] Jolly A., Estimation par télédétection satellitaire de la récolte an-nuelle en bois dans la futaie pure de pin maritime du massif desLandes de Gascogne, Apports pour la prévision de la ressourceforestière, Thèse de doctorat de l’Université Paul Sabatier deToulouse, Spécialité: Télédétection spatiale (N◦ 1626), 17 décem-bre 1993, 315 p.

[55] Jolly A., Guyon D., Riom J., Use of Landsat Thematic Mappermiddle infrared data to detect clearcuts in the Landes region, Int.J. Remote Sens. 17 (1996) 3615–3645.

[56] Joria P., Ahearn S., Connor M., A comparison of the SPOT andLandsat Thematic Mapper satellite systems for detecting gypsymoth defoliation in Michigan, Photogramm. Eng. Remote Sens.57 (1991) 1605–1612.

[57] Kasischke E.S., Bourgeau-Chavez L.L., French N.F., Observationsof variations in ERS-1 SAR image intensity associated with forestfires in Alaska, IEEE Trans. Geosci. Remote Sens. 32 (1994) 206–210.

[58] Kasischke E.S., Melack J.M., Dobson M.C., The Use of ImagingRadars for Ecological Applications – A Review, Remote Sens.Environ. 59 (1997) 141–156.

[59] Kerr Y.H., Lagouarde J.P., Imbernon J., Accurate land surface tem-perature retrieval from AVHRR data with use of an improved split-window algorithm, Remote Sens. Environ. 41 (1992) 197–209.

[60] Kimes B.L., Markham B.L., Tucker C.J., McMurtrey J.E.,Temporal Relationships Between Spectral Response andAgronomic Variables of a Corn Canopy, Remote Sens. Environ.11 (1981) 401–412.

[61] Kogan F.N., Drought of the late 1980s in the US as derived fromNOAA polar orbiting satellite data, Bull. Am. Meteorol. Soc. 76(1995) 655–667.

[62] Kuntz S., Siegert F., Monitoring of deforestation and land use inIndonesia with multitemporal ERS data, Int. J. Remote Sens. 20(1999) 2835–2853.

[63] Lagouarde J.-P., Ballans F., Moreau P., Guyon D., Coraboeuf D.,Experimental study of brightness surface angular variations ofmaritime pine (Pinus pinaster) stands, Remote Sens. Environ. 72(2000) 17–34.

[64] Leckie D.G., Jay C., Paradine D., Sturrock R., Preliminary as-sessment of Phellinus weirii infected (laminated root rot) treeswith high resolution CASI imagery, in: Hill D.A., Leckie D.G.(Eds.), Proceedings Automated Interpretation of High SpatialResolution Digital Imagery For Forestry, International Forum,Victoria, British Columbia, February 10–12, 1998, NaturalResources Canada, Canadian Forest Service, Pacific ForestryCentre, Victoria, British Columbia, 1999, pp. 187–195.

[65] Levesque J., King D.J., Airborne digital camera image semivari-ance for evaluation of forest structural damage at an acid minesite, Remote Sens. Environ. 68 (1999) 112–124.

[66] Liu W.T., Kogan F.N., Monitoring regional drought using theVegetation Condition, Int. J. Remote Sens. 17 (1996) 2761–2782.


[67] Lobo A., Maisongrande P., Stratified analysis of satellite imageryof SW Europe during summer 2003: the differential response ofvegetation classes to increased water deficit, Hydrol. Earth Syst.Sci. Discuss. 2 (2005) 2025–2060.

[68] López S., González F., Llop R., Cuevas M., An evaluation of theutility of NOAA-AVHRR images for monitoring forest fire risk inSpain, Int. J. Remote Sens. 12 (1991) 1841–1851.

[69] Luckman A., Baker J., Honzak M., Lucas R., Tropical forestbiomass density estimation using JERS-1 SAR: Seasonal varia-tion, confidence limits, and application to image mosaics – activeand passive, Remote Sens. Environ. 63 (1997) 126–139.

[70] Mårell A., Leitgeb E., European long-term research for sustainableforestry: experimental and monitoring assets at the ecosystem andlandscape level. Part 1: country reports, Technical Report 3, COSTAction E25, ECOFOR, Paris, 2005.

[71] Maselli F., Monitoring forest conditions in a protectedMediterranean coastal area by the analysis of multiyear NDVIdata, Remote Sens. Environ. 89 (2004) 423–433.

[72] Melia J., Lopez-Baeza E., Caselles V., Segarra D., Sobrino J.A.,Gilabert M.A., Moreno J., Coll C. EFEDA Annual Report, EPOC-CT90-0030 (LNBE), University of Valencia, 1991.

[73] Miller D.R., Quine C.P., Hadley W., An investigation of the poten-tial of digital photogrammetry to provide measurements of forestcharacteristics and abiotic damage, For. Ecol. Manage. 135 (2000)279–288.

[74] Myneni R.B., Keeling C.D., Tucker C.J., Asrar G., Nemani R.R.,Increased plant growth in the Northern high latitudes from 1981-1991, Nature, 386 (1997) 698–702.

[75] Ogée J., Brunet Y., Loustau D., Berbigier P., Delzon S., MuSICA,a CO2, water and energy multi-layer, multi-leaf pine forest model:evaluation from hourly to yearly time scales and sensitivity analy-sis, Glob. Change Biol. 9 (2003) 697–717.

[76] Pereira J.M.C., Govaerts Y., Potential Fire Applicationsfrom MSG/SEVIRI Observations. EUMETSAT ProgrammeDevelopment Department Technical Memorandum No. 07, 2001.

[77] Ranson K.J., Kovacs K., Sun G., Kharuk V.I., Disturbance recogni-tion in the boreal forest using radar and Landsat-7, Can. J. RemoteSens. 29 (2003) 271–285.

[78] Rederstorff D., Dépérissement forestier en Vallée du Rhin –Waldschäden im Rheintal. Photointerprétation du dépérissement.Mission aérienne du 9 août 1994, Office National des Forêts– Observatoire écologique de la Harth, Strasbourg, ForstlischeVersuch- und Forschunsanstalt Baden-Württemberg, Freiburg-im-Breisgau, 1996.

[79] Reichle R.H., Koster R.D., Dong J., Berg A.A., Global soil mois-ture from satellite observations, lands surface models and grounddata: implications for data assimilation, J. Hydrometeo. 5 (2004)430–442.

[80] Rigina O., Baklanov A., Hagner O., Olsson H., Monitoring of for-est damage in the Kola Peninsula, Northern Russia due to smeltingindustry, Sci. Total Environ. 229 (1999) 147–163.

[81] Rignot E., Salas W.A., Skole D.L., Mapping deforestation and sec-ondary growth in Rondonia, Brazil, using imaging radar and the-matic mapper data, Remote Sens. Environ. 59 (1997) 167–179.

[82] Riom J., Le dépérissement du Chêne - Apports de la Télédétection.1. Forêts des Pyrénées Atlantiques. 2. Forêt du Tronçais (Allier),in: Les Colloques de l’INRA, N◦ 32 : Applications de laTélédétection à l’Agriculture, INRA, Paris, 1984, pp. 117–145.

[83] Riom J., Apport de la télédétection pour l’estimation des dom-mages aux forêts vosgiennes, Colloque DEFORPA, Nancy, 24–26février 1988, 32 p.

[84] Riom J., Reteau F., Guyon D., Démonstration de l’application dela télédétection par méthode photographique pour l’estimation desdégâts aux forêts. Essai sur le massif forestier vosgien, C.R. defin de contrat CEE DG VI, APPF/II 5B (VIP F3/81) code INRA4100A, juin 1987, 120 p.

[85] Ripple W.J., Spectral reflectance relationships to leaf water stress,Photogramm, Eng. Remote Sens. 52 (1986) 1669–1775.

[86] Rouse J.W., Haas R.H., Schell J.A., Deering D.W., Monitoringvegetation systems in the Great Plains with ERTS, in: Freden S.C.,Becker M. (Eds.), Third earth resources technology satellite-1symposium, NASA SP-351, Washington, DC, 1974, Vol. 1,pp. 309–317.

[87] Saich P., Rees W.G., Borgeaud M., Detecting pollution damage toforests in the Kola peninsula using the ERS SAR, Remote Sens.Environ. 75 (2001) 22–28.

[88] Schwarz M., Steinmeier C., Holecz F., Stebler O., Wagner H.,Detection of windthrow in mountainous regions with different re-mote sensing data and classification methods, Scand. J. For. Res.18 (2003) 525–536.

[89] Seguin B., Becker F., Phulpin T., Gu X.F., Guyot G., Kerr Y., KingC., Lagouarde J.P., Ottlé C., Stoll P.M., Tabbagh A., Vidal A., Amini satellite project for land surface heat flux estimation fromfield to regional scale, Remote Sens. Environ. 68 (1999) 357–369.

[90] Siegert F., Kuntz S., Streck C., Bergbauer B., Land use plan-ning and monitoring in Indonesia using ERS-1 RADAR data,in: Proceedings International Conference on Remote Sensing andGIS for Environmental Resources Management – The Indonesian-European Experience, Jakarta, Indonesia, 6–8 June 1995.

[91] Skakun R.S., Wulder M.A., Franklin S.E., Sensitivity of theThematic Mapper enhanced wetness difference index to detectmountain pine beetle red-attack damage, Remote Sens. Environ.86 (2003) 433–443.

[92] Stach N., L’IFN cartographie les dégâts de la tempête sur le massifaquitain de pin maritime, Géomatique Expert 5 (2000) 15–17.

[93] Stach N., Deshayes M., Durrieu S., Mapping clear-cutting inFrench forests by satellite remote sensing, in: Olsson H. (Ed.),Proceedings of ForestSat 2005, Borås May 31–June 3 2005,Swedish National Board of Forestry 8a (2005) 118–125.

[94] Stach N., Deshayes M., Le Toan T., Mapping storm damage toforests using optical and radar remote sensing – The case of theDecember 1999 storms in France, in: Olsson H. (Ed.), Proceedingsof ForestSat 2005, Borås May 31–June 3 2005, Swedish NationalBoard of Forestry 8a (2005) 89–93.

[95] Stach N., Deshayes M., Le Toan T., Campillo D., Mathieu G.,Évaluation des dégâts de tempête par télédétection satellitaire : as-pects méthodologiques et opérationnels, Rapport final, ConventionDERF N◦ 2001-12-132, 2002

The contribution of remote sensing to the assessment of...

Documents

Transcript of The contribution of remote sensing to the assessment of...